이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 10 Sep 2018 02:06:40 +0000 (11:06 +0900)]
BUILD_TYPE=Release is specified in docker_env for tflite_benchmark_model (#2641)
Previously, build_type was specified in CI's groovy script.
But I would like to set all environment in one place - docker env file.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 7 Sep 2018 04:01:00 +0000 (13:01 +0900)]
Revise framework test list (#2636)
- Change directory structure: test/{tflite op name}/{testname}
- Add new framework tests to support more operators
- Update test list for neurun and benchmark_op
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
김용섭/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 7 Sep 2018 01:07:09 +0000 (10:07 +0900)]
Introduce EMBEDDING_LOOKUP op to pure_acl (#2631)
Introduce EMBEDDING_LOOKUP op to pure_acl
Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 7 Sep 2018 01:00:00 +0000 (10:00 +0900)]
[neurun] Implement Lowering operands (#2630)
This commit implements lowering for operands in `Graph::lower`.
operands lowering does the following:
- Normalize to 4D shape (rank greater than 4 does not supported)
- Collect info of operand layouts for def/use
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 6 Sep 2018 10:24:06 +0000 (19:24 +0900)]
[neurun] Introduce asShape4D (#2626)
This commit introduces asShape4D function to convert from
`operand::Shape` to `operand::LowerInfo::Shape4D`. This is
normalization to 4D shape.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 6 Sep 2018 10:23:56 +0000 (19:23 +0900)]
[neurun] Add LowerInfo to operand::Object (#2628)
Add LowerInfo to operand::Object as a member.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 6 Sep 2018 10:23:45 +0000 (19:23 +0900)]
Diff: initialize output tensor (#2627)
Initialize output tensor as zero
Need to RNN hidden state initialize
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 6 Sep 2018 09:48:48 +0000 (18:48 +0900)]
[neurun] Introduce operand::LowerInfo (#2622)
This commit introduces operand::LowerInfo which contains normalized
4D shape and tensor layout.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 6 Sep 2018 09:23:52 +0000 (18:23 +0900)]
Add Transpose into nnapi_delegate (#2624)
This commit enables TF Lite to call Transpose operation of NNAPI.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 6 Sep 2018 08:24:01 +0000 (17:24 +0900)]
[neurun] Add basic methods to LayoutSet (#2623)
Introduce `contains` and `size` method for user queries.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Enabling L2Pool2D Operation in NNAPI (#2587)
* Enabling PAD operation in NNAPI
This commit enables PAD operation in pure_arm_compute runtime
Signed-off-by: Siva Sai <siva.sai@samsung.com>
* Enabling L2Pool2D Operation in NNAPI
This commit enables L2Pool2D (Implicit Padding) operation in PACL runtime.
Related issue: #2562
Signed-off-by: Siva Sai <siva.sai@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 6 Sep 2018 05:34:56 +0000 (14:34 +0900)]
Remove unnecessary CAST_CL from pure_arm_compute (#2617)
This commit removes unnecessary CAST_CL from pure_arm_compute.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 6 Sep 2018 05:33:29 +0000 (14:33 +0900)]
Implement SPACE_TO_DEPTH operation for CPU (#2614)
This commit implements SPACE_TO_DEPTH operation in pure_arm_compute for CPU.
- Add a layer that supports SPACE_TO_DEPTH for CPU.
- Add a function to get ::arm_compute::Coordinates by the rank.
- Implement SPACE_TO_DEPTH operation.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 6 Sep 2018 04:17:09 +0000 (13:17 +0900)]
Use ExternalData for zero length value (#2618)
Use ExternalData for object with no value (nullptr buffer from setOperandValue)
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 6 Sep 2018 01:33:55 +0000 (10:33 +0900)]
TFlite file generator for FLOOR, PAD, SQUEEZE, TRANSPOSE (#2609)
This python script generates four tflite files for FLOOR, PAD, SQUEEZE and TRANSPOSE respectively.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 6 Sep 2018 01:33:22 +0000 (10:33 +0900)]
[neurun] Add IBackendConfig getter to Backend (#2611)
Add `IBackendConfig` getter `config()` method to `Backend`. We need to
keep it otherwise we do not have any way to call `getOperandLayout()`.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 6 Sep 2018 01:33:08 +0000 (10:33 +0900)]
[neurun] Introduce operand::LayoutSet (#2613)
LayoutSet is a set data structure for operand::Layout.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 6 Sep 2018 01:32:26 +0000 (10:32 +0900)]
[neurun] Move Source and Sink to `exec` dir (#2612)
Move `Source` and `Sink` to exec directory and change namespace from
`internal` to `exec`.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 6 Sep 2018 01:02:22 +0000 (10:02 +0900)]
[Model Parse] Parsing exceptional case: tensor index (#2593)
Parsing exceptional case when tensor index is less than 0
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 5 Sep 2018 10:43:34 +0000 (19:43 +0900)]
Introduce SPACE_TO_DEPTH operation into pure_arm_compute (#2601)
This commit introduce SPACE_TO_DEPTH operation into pure_arm_compute.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 5 Sep 2018 09:50:26 +0000 (18:50 +0900)]
[neurun] Change namespace of BackendXX to backend (#2603)
* [neurun] Change namespace of BackendXX to backend
Tshi commit changes namespace of BackendXX to backend.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
* Revert BackendResolver namespace
* Remove neurun::
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 5 Sep 2018 09:50:11 +0000 (18:50 +0900)]
Revert makfile to build pacl only (#2607)
Revert makfile to build pacl & test, not to build neurun
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 5 Sep 2018 09:05:57 +0000 (18:05 +0900)]
[neurun] Support iteration for operand::Object (#2608)
Support const/non-const iteration for operand::Object.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Wed, 5 Sep 2018 08:36:15 +0000 (17:36 +0900)]
[PureACL] Enable NEON to run Mobilenet (#2600)
This commit enables NEON to run Mobilenet by providing `depthwise_conv` and `reshape` for NEON.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 5 Sep 2018 08:32:28 +0000 (17:32 +0900)]
Add fullyconnected and mean tests using tflite (#2606)
This commit add fullyconnected and mean directory for testing these tflite files.
It will automatically download the files from npuarchive.mooo server and test it.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 5 Sep 2018 08:24:12 +0000 (17:24 +0900)]
[neurun] Enable compile option Wall/Wextra/Werror for backend modules (#2604)
This commit enables compile option Wall/Wextra/Werror for backend modules.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 5 Sep 2018 08:03:56 +0000 (17:03 +0900)]
[neurun] Use unordered_map instead of map (#2605)
Update `operand::Set` and `operation::Set` to use `std::unordered_map`
rather than `std::map`. Also update use key as Index instead of `int`.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 5 Sep 2018 05:42:08 +0000 (14:42 +0900)]
[neurun] Remove unused includes for BackendResolver (#2596)
Related #2510
This commit removes unused includes for BackendResolver.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 5 Sep 2018 05:41:56 +0000 (14:41 +0900)]
Define constructor for profile context (#2597)
Define constructor for profile context to initialize field
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 5 Sep 2018 05:35:51 +0000 (14:35 +0900)]
Initialize feature shape (#2598)
Initialize feature shape by constructor to remove uninitialized field
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 5 Sep 2018 05:06:58 +0000 (14:06 +0900)]
[neurun] Use Index as key for unordered_map (#2594)
This commit revises `unordered_map`s to use `graph::Index` as key
which used `int` as key because hash function was not exist.
Related PR : #2590
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 5 Sep 2018 04:18:02 +0000 (13:18 +0900)]
[neurun] Introduce getOperandLayout (#2591)
Introduce `IBackendConfig::getOperandLayout` which returns
`operand::Layout` for each backend. Note that this assumes a backend
only support one type of operand layout.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
SquaredDifference nnapi delegate and Custom ops addition (#2569)
This patch adds changes for SquaredDifference in nnapi delegate
and also adds related CustomOps changes.
SquaredDifference kernel has been added.
Signed-off-by: prasannar <prasanna.r@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 5 Sep 2018 04:15:24 +0000 (13:15 +0900)]
[neurun] Use Index as key for unordered_set (#2590)
This commit revises `unordered_set`s to use `graph::Index` as key
which used `int` as key because hash function was not exist.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 5 Sep 2018 04:14:12 +0000 (13:14 +0900)]
[neurun] Do not use addTensorBuilder (#2573)
PR #2563 has introduced `IPlanBuilder::addTensorBuilder`. However
adding a call to `addTensorBuilder` for all nodes from Planner is an
overhead for development. So this commit revises to remove the calls.
However with this commit, `Linear::markTensor` method returns
`TensorBuilderSet` and passes it to `PlanBuilder::finalize`. Still
need to enhance this structure, so TODO comments are left.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 5 Sep 2018 03:15:55 +0000 (12:15 +0900)]
[Model Parser] Update tflite file API (#2580)
Update tflite file API: schema in tensorflow r1.10
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 5 Sep 2018 02:17:40 +0000 (11:17 +0900)]
[neurun] Rename IBackendInitializer (#2583)
Rename IBackendInitializer to IBackendConfig. This is for adding other
methods to this interface class.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Wed, 5 Sep 2018 01:46:01 +0000 (10:46 +0900)]
Introduce `benchmark_tflite_model` option in test_driver.sh (#2582)
Related Issue: #2544
In addition, garden test_driver.sh
- Align the usage message
- $MY_PATH is renamed to $TEST_DRIVER_DIR
- $ARTIFACT_PATH is not mandatory option any more.
It is set based on $TEST_DRIVER_PATH automatically
`run_benchmark_tflite_model.sh` will be added in next PR.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Wed, 5 Sep 2018 01:07:36 +0000 (10:07 +0900)]
Add docker_env_pureacl_tflite_benchmark_model (#2566)
To build tflite_benchmark_model using docker, the corresponding docker
environment is added.
To distinguish tflite benchmark model for neurun, the env file gets
pureacl in the middle of the name.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 5 Sep 2018 01:02:28 +0000 (10:02 +0900)]
[neurun] Specialize `std::hash` for Index class (#2585)
(partially) Specialize `std::hash` for `graph::Index` class.
This will enable having Index as the key in "std::unordered_.."
containers.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 5 Sep 2018 00:59:23 +0000 (09:59 +0900)]
[Model Parser] Update flatbuffers path (#2579)
Update flatbuffers path to use flatbuffer python library
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 4 Sep 2018 23:31:21 +0000 (08:31 +0900)]
Catch exception in nnapi test (#2581)
Catch exception in nnapi test to avoid warning
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 4 Sep 2018 09:51:49 +0000 (18:51 +0900)]
Use unique pointer for ANeuralNetworksMemory (#2578)
Use unique pointer for ANeuralNetworksMemory to avoid memory leak warning
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 4 Sep 2018 08:39:09 +0000 (17:39 +0900)]
[neurun] Change path for wrapping structure (#2571)
Change path for NeuralNetworks.h's wrapping structure
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 4 Sep 2018 08:16:54 +0000 (17:16 +0900)]
Coverity cross build & analyze script (#2560)
Coverity cross build & analyze script for daily coverity report
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Removing GeneratedTests.pad from the skiplist (#2574)
As #2097 is merged, this commit removes GeneratedTests.pad from the skiplist.
Signed-off-by: Siva Sai <siva.sai@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 4 Sep 2018 06:03:23 +0000 (15:03 +0900)]
[neurun] Do BackendResolver in lowering (#2563)
* [neurun] Do BackendResolver in lowering
This commit moves doing BackendResolver into lowring in graph.
- TensorBuilder doesn't get Plan in constructor anymore.
- LowerInfo includes Backend class not backend string.
- BackendResolver returns only Backend class.
Signed-off-by: sjsujinkim sjsujin.kim@samsung.com
* Change backend() to getBackend()
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 4 Sep 2018 04:54:21 +0000 (13:54 +0900)]
Initialize members in DiffSummary constructor (#2568)
Initialize members in DiffSummary constructor to avoid warning
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 4 Sep 2018 04:54:08 +0000 (13:54 +0900)]
[neurun] Define default value in constructor (#2567)
Define default value in constructor for kernel
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 4 Sep 2018 04:07:38 +0000 (13:07 +0900)]
[neurun] Enhance DFS implementation (#2561)
Now that we have operand def/use info, we can do DFS efficiently.
This is applied to PostDfsIterator. This might also be applied to
DAGVerifier, but operand def/use info is built when Graph building is
finished.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 4 Sep 2018 03:56:43 +0000 (12:56 +0900)]
[neurun] Use MockNode.h in Insert test (#2565)
This commit uses `MockNode.h` in Insert test.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
nnfw Adding SquaredDifferenceEx in NeuralNetworksEx.h (#2555)
This patch adds '''ANEURALNETWORKS_SQUARED_DIFFERENCE_EX''' in the ''' NeuralNetworksEx.h'''
Signed-off-by: prasannar <prasanna.r@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 3 Sep 2018 11:28:37 +0000 (20:28 +0900)]
[neurun] Use forward declaration from a header (#2557)
Forward declare for the interfaces that are used from
`BackendManger.h`.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 3 Sep 2018 11:04:30 +0000 (20:04 +0900)]
[neurun] Use forward decl from Node (#2558)
Forward declare `LowerInfo` from `operation::Node` to reduce
dependencies.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 3 Sep 2018 10:34:13 +0000 (19:34 +0900)]
[neurun] Change check condition for addOperation & addOperationEx (#2551)
Change check condition for addOperation & addOperationEx to same with pacl
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Enabling PAD operation in NNAPI (#2097)
This commit enables PAD operation in pure_arm_compute runtime
Signed-off-by: Siva Sai <siva.sai@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 3 Sep 2018 07:10:56 +0000 (16:10 +0900)]
[neurun] Range-based loop for operand::IndexSet (#2554)
Support and apply range-based loop for operand::IndexSet and remove
`list()` method.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 3 Sep 2018 05:44:29 +0000 (14:44 +0900)]
[neurun] Move CLSchedule default init to BackendManager (#2526)
* [neurun] Move CLSchedule default init to BackendManager
This commit moves CLSchedule default init to BackendManager, since it is ACL_CL specific code.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
* Extract backend initializer
* Add eol
* Remove unnecessary variable
* Remove unused include
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 3 Sep 2018 05:42:36 +0000 (14:42 +0900)]
Change check condition for addOperationEx (#2550)
Change check condition for addOperationEx to same with addOperation
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 3 Sep 2018 05:04:43 +0000 (14:04 +0900)]
[tool] OpenCL info (#2549)
This tool shows information from OpenCL : info of paltform, context and devices.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 3 Sep 2018 04:44:17 +0000 (13:44 +0900)]
[neurun] Return nullptr for NOP stage generator (#2553)
Return nullptr for NOP stage generator
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 3 Sep 2018 04:43:18 +0000 (13:43 +0900)]
[neurun] Comment out unreachable code (#2552)
Comment out unreachable code in cpu kernel to avoid svace warning
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 3 Sep 2018 02:31:00 +0000 (11:31 +0900)]
[neurun] Update use-def init test (#2545)
Use SimpleMockNode in header
Remove unused namespace alias
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 3 Sep 2018 01:11:37 +0000 (10:11 +0900)]
[neurun] Make initializeUseDef less verbose (#2546)
Make `Graph::initializeUseDef` implementation less verbose via using
directly a lambda function.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 31 Aug 2018 08:29:17 +0000 (17:29 +0900)]
[neurun] Introduce DefaultIterator (#2535)
Introduce a default iterator that just iterates in the order of
`operation::Set::iterate` method.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 31 Aug 2018 08:28:44 +0000 (17:28 +0900)]
Add memcpy() version of generic TensorSource for 3D tensor (#2477)
This commit adds memcpy() version of generic TensorSource for 3D tensor.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 31 Aug 2018 08:22:05 +0000 (17:22 +0900)]
Add a default param to asTensorShape() to apply conditionally (#2464)
This commit add a default param to asTensorShape() to apply conditionally dimension coorection.
In some cases, in incorrect dimensions is required.
For example, intput_size is 1 in LSTM. The input-to-input weights([num_units, input_size]) of LSTM is used as the weight of the FullyConnected.
The FullyConnected's weight must be greater or equal than 2-dimensions.
However, if the dimension correction is applied to input_to_input_weights with input_size equal to 1, it will be changed to 1-D.
So input_to_input_weights is not used by the weight of FullyConnected.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 31 Aug 2018 07:34:09 +0000 (16:34 +0900)]
[neurun] Initialize use & def information (#2543)
Initialize use & def information in model finishing phase
Add test code to check initialized use&def information
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 31 Aug 2018 06:54:50 +0000 (15:54 +0900)]
Add sourcing tensor more than 5 dimensions (#2529)
This commit adds sourcing tensor more than 5 dimensions.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 31 Aug 2018 06:51:43 +0000 (15:51 +0900)]
[neurun] Introduce SimpleMockNode for Graph test (#2536)
Introduce SimpleMockNode as a header file so it can be reused from
many tests.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 31 Aug 2018 06:21:49 +0000 (15:21 +0900)]
[neurun] Define use & def in operand (#2541)
- Define use & def in operand
- Define base methods to append & remove
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 31 Aug 2018 03:55:34 +0000 (12:55 +0900)]
Add assertion in ToARMCompute function (#2540)
This commit prevents a segfault from wrong access.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 30 Aug 2018 22:23:31 +0000 (07:23 +0900)]
Unify to use one function of asTensorShape() and asTensorInfo() (#2471)
* Unify to use one function of asTensorShape() and asTensorInfo()
This commit unifies to use one function(no overloading) of asTensorShape() and asTensorInfo().
- Export asTensorShape() out of asTensorInfo() to apply conditionally dimension correction.
- Unify to use one function to eliminate unnecessary complexity.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
* Remove unused overloading functions
This commit remove unused overloading functions.
- asTensorInfo()
- asTensorShape()
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 30 Aug 2018 09:47:46 +0000 (18:47 +0900)]
Update tensorflow register file (#2531)
These operators are supported by Tensorflow v1.9.0
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 30 Aug 2018 08:13:18 +0000 (17:13 +0900)]
Return nullptr for InitRSQRT (#2533)
Return nullptr for InitRSQRT to avoid svace warning
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 30 Aug 2018 08:12:57 +0000 (17:12 +0900)]
[neurun] Remove remain common tensor codes (#2528)
This commit removes remain common tensor codes.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 30 Aug 2018 08:12:41 +0000 (17:12 +0900)]
[neurun] Move Op.lst to graph/operation (#2530)
Move the file Op.lst to graph/operation from internal/op.
Now we no longer have any files in internal/op.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 30 Aug 2018 08:12:03 +0000 (17:12 +0900)]
[neurun] Define operation index list (#2532)
Define operation index list (used for use-def)
Use list to support frequent append & remove
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 30 Aug 2018 07:51:27 +0000 (16:51 +0900)]
[neurun] Enable compile option Wall/Wextra/Werror (#2517)
To keep our code clean, let us introduce those compile options.
- Wall : Enable most(not exactly all) of compiler warnings
- Wextra : Enable extra compiler warnings
- Werror : Treat warnings as error
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 30 Aug 2018 07:02:30 +0000 (16:02 +0900)]
[neurun] Move internal/operand to graph/operand (#2520)
Move internal/operand to graph/operand
Change namespace neurun::internal::operand to neurun::graph::operand
Remove internal/operand
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
김용섭/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 30 Aug 2018 06:58:58 +0000 (15:58 +0900)]
Remove all fc tests of generated tests from skiplist (#2527)
- V1_0/fully_connected_float_1_nnfw
- V1_0/fully_connected_quant8_2
- V1_1/fully_connected_float_4d_simple
Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
김용섭/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 30 Aug 2018 05:00:04 +0000 (14:00 +0900)]
Introduce GenericFullyConnectedLayer for fc (#2525)
4d fc tests of GeneratedTests are failed because runtimes can't
guess 4d input to 2d so far. This GenericFullyConnectedLayer handles the
case.
Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 30 Aug 2018 01:41:31 +0000 (10:41 +0900)]
Enable concat-related generatedTests (#2523)
All concat generatedTests work normally after merging #2494
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 30 Aug 2018 01:02:46 +0000 (10:02 +0900)]
[neurun] Remove internal::op nodes (#2519)
Remove Nodes in the namespace internal::op and some classes related to
those nodes.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 29 Aug 2018 09:01:43 +0000 (18:01 +0900)]
[neurun] Fix build break (#2521)
Fix build break which is due to invalid shape_index of Reshape
operation.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 29 Aug 2018 07:55:21 +0000 (16:55 +0900)]
[neurun] Remove TensorMarker (#2515)
Now that we have interface `getInputs()` and `getOutputs()` in
operation Node, we do not need to implement TensorMarker for each
Node.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 29 Aug 2018 07:54:03 +0000 (16:54 +0900)]
[neurun] Fix Reshape operation (#2516)
shape_index is a tensor paramenter so it should be one of the inputs.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 29 Aug 2018 07:49:46 +0000 (16:49 +0900)]
Update runtime build in Makefile (#2518)
Update runtime build in Makefile to build neurun with purecl
It will be used for svace
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 29 Aug 2018 07:37:42 +0000 (16:37 +0900)]
Support the broadcast of all cases (#2511)
* Support the broadcast of all cases
This commit supports the broadcast of all case.
- Add Extending the rank of operand.
- Apply the bradcast to the arithmetic operations.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
* Remove unnecessary residual codes related to the broadcast
This commit removes unnecessary residual codes related to the broadcast.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 29 Aug 2018 07:12:17 +0000 (16:12 +0900)]
Support CONCAT for all dimensions and all axes (#2494)
This commit is similar with SPLIT implementation.
In order to avoid unnecessary copy around data,
it uses shared memory information.
It does not u se separate ACL function like SPLIT.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
nnfw: RSQRT nnapi delegate and Custom ops addition. (#2472)
* nnfw: RSQRT nnapi delegate and Custom ops addition.
These patch adds changes for RSQRT in nnapi_delegate
and also adds related CustomOps changes.
RSQRT kernel has been added.
Signed-off-by: Tanuj Tekriwal <tanuj.tekri@samsung.com>
* nnfw: Add pure_arm_compute layer for RSQRT
This patch adds the pure_arm_compute layer for RSQRT.
Related ACL changes to be added in different patch.
Signed-off-by: Tanuj Tekriwal <tanuj.tekri@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 29 Aug 2018 05:41:05 +0000 (14:41 +0900)]
[neurun] Update Dumper class (#2513)
As we have unified graph operation nodes, this commit updates class
`Dumper` accordingly.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 29 Aug 2018 02:29:59 +0000 (11:29 +0900)]
Add an initializer for 3D tensor (#2487)
This commit adds initializer for 3D tensor.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
남궁석/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 29 Aug 2018 02:16:27 +0000 (11:16 +0900)]
Enable to check state of compilation in pureacl (#2439)
This commit will enable to check if compilation is finished or not
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 29 Aug 2018 01:45:03 +0000 (10:45 +0900)]
[neurun] Merge graph node classes (#2497)
Merge Old Node into New Node.
Old Node : `internal::tflite::op::Node`
New Node : `neurun::graph::operation::Node`
- struct `Param` for each Node holds non-tensor parameters only
- Tensor parameters are stored in `_inputs` and `_outputs`
- Provide a default implementation of get/setters for inputs and
outputs of `Node`
- `NodeVisitors` are now for New Node, not Old Node
- class `Linear` now holds New Node rather than Old Node
This commit completes the draft commit #2479, with #2483 and #2486
which were already merged.
Resolve #2404
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 29 Aug 2018 01:44:28 +0000 (10:44 +0900)]
[neurun] Move CPU Tensor implementation to backend (#2509)
Move `internal/cpu.{h|cc}` to cpu backend directory.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 29 Aug 2018 01:33:33 +0000 (10:33 +0900)]
Fix axis calculation of TensorSource from reversing to using ToARMComputeAxis (#2481)
This commit Fix axis calculation of TensorSource from reversing to using ToARMComputeAxis.
Supported tensor rank of TensorSource:
1D ~ 3D : reversing
more than 4D : ...NHWC -> WHCN...
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 29 Aug 2018 01:31:11 +0000 (10:31 +0900)]
Update arm ComputeLibrary version (#2508)
Update ComputeLibrary to rebased version instead of merged one.
Note that this has zero diff, but removed merge commit.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 29 Aug 2018 00:49:48 +0000 (09:49 +0900)]
[neurun] Add insertOperation in graph (#2424)
* [neurun] Add insertOperation in graph
This commit adds insertOperation which inserts between 2 operations in graph.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
* some change test operand
* remove unused declaration
* get a pair of operand and operation index as parameter
* Add todo comment into insertOperation
* Change pair to seperate params
* Revise insertion test
* fix format
* Remove unused utility include
* rebase
* Change test comment
* Remove test header
김용섭/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 28 Aug 2018 10:34:36 +0000 (19:34 +0900)]
Prepare reshaping input(rank-4) into rank-2 on fc (#2505)
When input which has rank 4 as fc's input comes, it needs to be reshaped
into rank 2. This commit prepares the task by skeleton code and checking
input_size, output_size and batch_size
Signed-off-by: Yongseop Kim <yons.kim@samsung.com>